Target detection in latent space

1. A method for processing medical image data, comprising:

inputting medical image data to a variational autoencoder configured to reduce a dimensionality of the medical image data to a latent space having one or more latent variables with latent variable values as an encoder, such that the latent variable values corresponding to an image with no tissue of a target tissue type fit within one or more clusters, the variational autoencoder having been trained with the encoder and a decoder configured to regenerate images from the training values of the latent variables output by the encoder;

determining a probability that the latent variable values corresponding to the medical image data fit within the one or more clusters of latent variable values of training image data for the variational autoencoder, the probability determined from the latent variable values for the medical image output by the encoder relative to the clusters of the latent variable values of the training image data for the variational autoencoder; and

determining that a tissue of the target tissue type is present in response to a determination that the medical image data have less than a threshold probability of fitting within any of the one or more clusters based on the latent variable values.

2. The method of claim 1 , wherein determining the probability includes performing an outlier detection process.

3. The method of claim 2 , wherein the outlier detection process includes a random sampling consensus method.

4. The method of claim 2 , wherein the outlier detection process includes a random forest method.

5. The method of claim 2 , wherein the outlier detection process includes performing a statistical test.

6. The method of claim 2 , wherein the outlier detection process includes a classification method.

7. The method of claim 1 , further comprising, before the inputting, training the variational autoencoder using a set of the training image data representing images from subjects without the target tissue type.

8. The method of claim 7 , wherein the training includes deep unsupervised learning.

9. The method of claim 7 , further comprising:

clustering the training image data into the one or more clusters in the latent variables based on values of an input parameter; and

determining a probability that a subject having a subject input parameter value also has a latent variable value that fits within one of the one or more clusters corresponding to the subject input parameter value.

10. A medical image system, comprising:

a non-transitory, machine readable storage medium storing program instructions and medical image data; and

a programmed processor coupled to the storage medium and configured by the program instructions for:

input of medical image data to a variational autoencoder configured to reduce a dimensionality of the medical image data to a latent space having one or more latent variables with latent variable values as an encoder, such that the latent variable values corresponding to an image with no tissue of a target type fit within one or more clusters of the values of the latent variables, the variational autoencoder having been trained with the encoder and a decoder configured to regenerate images from the training values of the latent variables output by the encoder;

detection of whether the latent variable values corresponding to the medical image data fit within the one or more clusters of latent variable values of training image data for the variational autoencoder, the fit determined from the latent variable values for the medical image output by the encoder relative to the clusters of the latent variable values of the training image data for the variational autoencoder; and

determination that a tissue abnormality is present in response to a determination that the medical image data have less than a threshold probability of fitting within any of the one or more clusters.

11. The medical image system of claim 10 , wherein the detection includes performance of an outlier detection.

12. The medical image system of claim 11 , wherein the outlier detection includes a random sampling consensus.

13. The medical image system of claim 10 , wherein the medical image data comprises magnetic resonance (MR) data, and the system further comprises an MR scanner coupled to the processor and the storage medium, the MR scanner configured to acquire MR signals for reconstructing the MR image data.

14. The medical image system of claim 10 , wherein the medical image data comprises magnetic resonance (MR) data, and the processor is configured to receive MR signals from an MR scanner and reconstruct the MR image data.

15. The medical image system of claim 10 , further comprising a display coupled to the processor, wherein the processor is configured for:

display of a plot of a set of latent variable values corresponding to a set of training images, where the training images have no tissue of the target type, the plot further including the latent variable values corresponding to the medical image data.

16. The medical image system of claim 10 , wherein the processor is configured to cause the latent variable values corresponding to the medical image data to be displayed differently from the set of latent variable values corresponding to a set of training images.